参考文献:
iwslt2017-ja-it
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-it')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1566年 |
'train' | 231619 |
'validation' | 929 |
- 特徴:
{
"translation": {
"languages": [
"en",
"it"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-nl
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-nl')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1777年 |
'train' | 237240 |
'validation' | 1003 |
- 特徴:
{
"translation": {
"languages": [
"en",
"nl"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-en-ro
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-ro')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1678年 |
'train' | 220538 |
'validation' | 914 |
- 特徴:
{
"translation": {
"languages": [
"en",
"ro"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-it-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-it-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1566年 |
'train' | 231619 |
'validation' | 929 |
- 特徴:
{
"translation": {
"languages": [
"it",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-it-nl
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-it-nl')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1669年 |
'train' | 233415 |
'validation' | 1001 |
- 特徴:
{
"translation": {
"languages": [
"it",
"nl"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-it-ro
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-it-ro')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1643年 |
'train' | 217551 |
'validation' | 914 |
- 特徴:
{
"translation": {
"languages": [
"it",
"ro"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-nl-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-nl-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1777年 |
'train' | 237240 |
'validation' | 1003 |
- 特徴:
{
"translation": {
"languages": [
"nl",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-nl-it
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-nl-it')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1669年 |
'train' | 233415 |
'validation' | 1001 |
- 特徴:
{
"translation": {
"languages": [
"nl",
"it"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-nl-ro
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-nl-ro')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1680年 |
'train' | 206920 |
'validation' | 913 |
- 特徴:
{
"translation": {
"languages": [
"nl",
"ro"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ro-en
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ro-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1678年 |
'train' | 220538 |
'validation' | 914 |
- 特徴:
{
"translation": {
"languages": [
"ro",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ro-it
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ro-it')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1643年 |
'train' | 217551 |
'validation' | 914 |
- 特徴:
{
"translation": {
"languages": [
"ro",
"it"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ro-nl
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ro-nl')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 1680年 |
'train' | 206920 |
'validation' | 913 |
- 特徴:
{
"translation": {
"languages": [
"ro",
"nl"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ar-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ar-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8583 |
'train' | 231713 |
'validation' | 888 |
- 特徴:
{
"translation": {
"languages": [
"ar",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-de-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8079 |
'train' | 206112 |
'validation' | 888 |
- 特徴:
{
"translation": {
"languages": [
"de",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-ar
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-ar')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8583 |
'train' | 231713 |
'validation' | 888 |
- 特徴:
{
"translation": {
"languages": [
"en",
"ar"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-de
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-de')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8079 |
'train' | 206112 |
'validation' | 888 |
- 特徴:
{
"translation": {
"languages": [
"en",
"de"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-fr
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-fr')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8597 |
'train' | 232825 |
'validation' | 890 |
- 特徴:
{
"translation": {
"languages": [
"en",
"fr"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-ja')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8469 |
'train' | 223108 |
'validation' | 871 |
- 特徴:
{
"translation": {
"languages": [
"en",
"ja"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-en-ko
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-ko')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8514 |
'train' | 230240 |
'validation' | 879 |
- 特徴:
{
"translation": {
"languages": [
"en",
"ko"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-zh
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-en-zh')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8549 |
'train' | 231266 |
'validation' | 879 |
- 特徴:
{
"translation": {
"languages": [
"en",
"zh"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-fr-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-fr-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8597 |
'train' | 232825 |
'validation' | 890 |
- 特徴:
{
"translation": {
"languages": [
"fr",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ja-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ja-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8469 |
'train' | 223108 |
'validation' | 871 |
- 特徴:
{
"translation": {
"languages": [
"ja",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-ko-en
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-ko-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8514 |
'train' | 230240 |
'validation' | 879 |
- 特徴:
{
"translation": {
"languages": [
"ko",
"en"
],
"id": null,
"_type": "Translation"
}
}
iwslt2017-zh-ja
次のコマンドを使用して、このデータセットを TFDS にロードします。
ds = tfds.load('huggingface:iwslt2017/iwslt2017-zh-en')
- 説明:
The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task. The languages involved are five:
German, English, Italian, Dutch, Romanian.
For each language pair, training and development sets are available through the entry of the table below: by clicking, an archive will be downloaded which contains the sets and a README file. Numbers in the table refer to millions of units (untokenized words) of the target side of all parallel training sets.
- ライセンス: 不明なライセンス
- バージョン: 1.0.0
- 分割:
スプリット | 例 |
---|---|
'test' | 8549 |
'train' | 231266 |
'validation' | 879 |
- 特徴:
{
"translation": {
"languages": [
"zh",
"en"
],
"id": null,
"_type": "Translation"
}
}